HiFi-GAN

To set up the HiFi-GAN environment, ensure you meet the following requirements:

1. Python: Version 3.6 or higher.
2. Repository: Clone the HiFi-GAN repository.
3. Dependencies: Install required Python packages listed in requirements.txt.
4. Dataset: Download and extract the LJ Speech dataset and move all .wav files to the LJSpeech-1.1/wavs directory.

To generate audio directly from mel-spectrograms (e.g., from Tacotron2 or Glow-TTS):

1. Create a directory named test_mel_files and copy your generated mel-spectrogram files into it.
2. Run inference_e2e.py providing the path to a generator checkpoint.

Generated audio is saved to generated_files_from_mel by default. Use the --output_dir option to specify a different destination.

python inference_e2e.py --checkpoint_file [generator checkpoint file path]

You can train the HiFi-GAN generator using the train.py script. By default, checkpoints and configuration copies are saved in the cp_hifigan directory.

To train different versions (V1, V2, or V3), select the corresponding configuration file (config_v1.json, config_v2.json, or config_v3.json).

Use the --checkpoint_path option to specify a custom directory for saving checkpoints.

python train.py --config config_v1.json

Fine-tuning allows you to adapt the model to new data using mel-spectrograms.

Steps:

1. Prepare Mel-spectrograms: Generate mel-spectrograms in .npy format (using a tool like Tacotron2 with teacher-forcing). The filename must match the original audio file (e.g., LJ001-0001.wav $\rightarrow$ LJ001-0001.npy).
2. Organize Data: Create a directory named ft_dataset and copy all .npy files into it.
3. Execute Training: Run train.py with the --fine_tuning flag set to True.

python train.py --fine_tuning True --config config_v1.json

To reconstruct audio from existing .wav files (mel-spectrogram inversion):

1. Create a directory named test_files and copy your target .wav files into it.
2. Run inference.py providing the path to a generator checkpoint.

Generated audio is saved to generated_files by default. Use the --output_dir option to specify a different destination.

python inference.py --checkpoint_file [generator checkpoint file path]

The following pretrained models are available for use: